SK Group, TSMC Deepen AI Chip Partnership to Develop Next-Generation HBM4 Technology

High-bandwidth memory is a critical component in AI accelerators, providing the data throughput required by advanced processors. HBM4E is the next step in that product cycle, and its base logic die requires sophisticated foundry manufacturing. Adding a second supplier could help SK Hynix contain rising process costs and reduce dependence on a single manufacturing partner, while challenging TSMC’s reported exclusive position in the company’s outsourced HBM logic-die production.

SK Hynix is evaluating whether to allocate part of its HBM4E base-die production to Intel Foundry, according to the latest report, creating a dual-source arrangement alongside TSMC. As of Aug. 31, 2026, the plan remained under consideration, with no disclosed production schedule, process node, wafer volume or contract value. A deal would give Intel a foothold in the fast-growing AI memory supply chain and could weaken TSMC’s exclusive role in the program.

High-bandwidth memory, or HBM, is a critical component that gives AI accelerators rapid access to vast amounts of data, directly affecting model training and inference efficiency. SK hynix is targeting next-generation AI data centers and large-scale computing systems with HBM4E, while using MR-MUF packaging to improve cooling, performance and stability in multilayer stacks.

SK hynix recently said it had supplied 12-layer HBM4E samples to major customers but did not disclose their names, shipment dates or transaction values. The new product delivers data-transfer speeds of 16 Gbps per pin and more than 20% greater energy efficiency than its predecessor. The company hopes customer validation will accelerate adoption in next-generation AI chips.

AI factories depend on high-bandwidth, energy-efficient memory, making performance and development cycles critical to scaling AI computing. Longtime partners NVIDIA and South Korea’s SK hynix are expanding their relationship beyond the supply of products such as HBM into joint design for the AI infrastructure, personal AI and physical AI markets.

NVIDIA and SK hynix announced a multiyear partnership on June 8, 2026, without disclosing its value. The companies will develop dedicated memory for the Vera Rubin AI supercomputer, Vera CPU, RTX Spark PCs and Jetson Thor platform. SK hynix will also adopt NVIDIA’s AI and digital twin technologies to accelerate chip design, manufacturing and decision-making.

High-bandwidth memory (HBM) delivers large volumes of data to AI accelerators with high bandwidth and low power consumption, making it a critical component in the race for generative AI computing power. After a downturn in the memory market, SK hynix rebounded through technology investment and its supply partnership with Nvidia, becoming a key company in the global AI semiconductor supply chain.

South Korean broadcaster KBS was granted access to SK hynix’s M16 wafer fabrication plant for the first time. Its program “The World Seen for the First Time” showed the company’s secret HBM production line operating around the clock, along with employees at work. The report did not specify an air date, investment amount or production capacity, but footage from the facility showed the company working at full speed to meet demand for AI memory.

High-bandwidth memory, or HBM, is a critical component that gives AI accelerators rapid access to data, and supply has fallen short as data centers expand worldwide. SK hynix currently controls about 60% of the HBM market and is using advanced packaging and next-generation products to strengthen its lead while competing with Samsung Electronics and Micron for orders from major cloud providers.

SK Group Chairman Chey Tae-won recently pledged to sharply increase SK hynix's HBM output and accelerate the latest HBM4 into mass production to narrow the supply-demand gap for AI chips. He said adding wafer capacity takes at least four years and that tight global supplies could persist until 2030. The reports did not disclose the investment amount for the expansion or a specific production start date.